Notes

• Integrating LLM agents with edge computing creates MEGI for real-time, private reasoning.
• Deploying LLMs in MEGI faces high compute demands and limited edge resources.
• A joint framework combines adaptive CoT reasoning with distributed MoE deployment.
• Innovation: Jointly optimizing reasoning depth, expert activation, and transmission power.
• The system dynamically adjusts expert networks and complexity per task and device.
• Experiments confirm the framework balances quality and efficiency, proving MEGI viability.

• LLM与边缘计算结合催生了MEGI，提供实时、注重隐私的边缘推理。
• MEGI面临LLM高算力需求与边缘设备资源有限的挑战。
• 提出联合优化框架，结合自适应CoT推理与分布式MoE部署。
• 核心创新：将推理深度视为动态资源，与专家激活及传输功率联合优化。
• 系统按任务与设备能力，动态调节专家网络与推理复杂度。
• 实验证明该框架平衡了推理质量与资源，验证了其在MEGI中的可行性。